Preprint in preparation · IGVF Consortium · UCLA

Epigenomic heterogeneity and spatial organization during human reprogramming

Terence W. Li1,2,3,*, Justin Langerman4,*, Cuining Liu1,3,*, Yu Sun4,*, Mohammad S. Baig1,5, Gianna Kim1, Jingyuan Fu4,6, Rayyan Irfan Ghoor1, Yi Zhang1, Zach Von Behren1, Min Jen Tsai1, Malina Elena Cantemir1, Kevin D. Abuhanna1,5, Edward Chen1, Jon Paino6, Gideon Shaked4, Lily Zhao4, Lisa Barooah4, Amy Sun4, Andrea Garcia Angulo4, Alexander Lee4, Sophia Peavy4, Eleazar Eskin1,2,6, Noah Zaitlen1,2,7, Brunilda Balliu2,8,9,†, Jason Ernst2,4,6,†, Chongyuan Luo1,†, Kathrin Plath4,†

Affiliations · *co-first authors · co-corresponding authors
  1. Department of Human Genetics, David Geffen School of Medicine, UCLA
  2. Department of Computational Medicine, David Geffen School of Medicine, UCLA
  3. Bioinformatics Interdepartmental Program, UCLA
  4. Department of Biological Chemistry, David Geffen School of Medicine, UCLA
  5. Genetics and Genomics Graduate Program, UCLA
  6. Computer Science Department, Henry Samueli School of Engineering and Applied Science, UCLA
  7. Department of Neurology, David Geffen School of Medicine, UCLA
  8. Department of Pathology and Laboratory Medicine, David Geffen School of Medicine, UCLA
  9. Department of Biostatistics, Fielding School of Public Health, UCLA

Summary

Reprogramming somatic cells to pluripotency requires extensive remodeling of transcriptional and epigenetic states, yet most cells never complete the transition. We integrated single-cell gene expression, chromatin accessibility, DNA methylation and three-dimensional genome organization across densely sampled reprogramming time courses of fibroblasts from four individuals, together with highly multiplexed spatial transcriptomics.

  1. 1DNA methylation carries individual identity that other layers miss. Starting fibroblasts show extensive inter- and intra-individual methylation heterogeneity — concentrated in partially methylated domains — that is far less apparent in RNA, ATAC or 3D contacts, and diminishes only along the productive trajectory.
  2. 2The two failure branches fail differently. Cells diverging early acquire inflammatory and antiviral transcription while their methylation and 3D-genome changes revert toward the fibroblast state. Cells that stall later advance transcriptionally but leave methylation remodeling incomplete.
  3. 3An early-iPSC transcriptome is not one endpoint. Cells reaching a similar early iPSC transcriptional state separate into two distinct methylation and 3D-genome configurations, one far closer to mature iPSCs — with 80.9% of differentially methylated regions overlapping differential TAD boundaries.
  4. 4Fate is spatially organized by day 3. In situ profiling of 139,407 cells shows productive cells already concentrated in hotspots at the earliest sampled stage, with fewer hotspots in the low-efficiency donor.

Figure 1

Click the figure for the full-resolution PNG.

Data

Raw/processed data is deposited on the IGVF portal and publicly released. The five principal analysis sets below are post-QC, and we provide download links for all raw/intermediate files used to generate them.

Raw measurement sets — FASTQ + seqspec Processed intermediate analysis sets — alignments, matrices, fragments Analysis-ready principal analysis sets — final matrices + cell annotations

Which donor is in which assay

The portal identifies donors by accession; the manuscript calls them C29, C37, C38 and C39.

Download

Direct links, one file at a time or a whole tier at once. Every link points at api.data.igvf.org/…/@@download/…, which redirects to public S3 — it works in a browser, in curl, and on a cluster with no credentials.

Start here — analysis-ready files

The final per-modality outputs: . Grouped by file type, one click per file.

Processed and raw files, by type

Whole-tier pulls. Each row gives you a manifest — a TSV with one row per file including its download URL, size and md5 — either live from the portal or as a static copy shipped with this page.

Build a download set

Pick a modality, tier, file types and time points; get back a URL list or a ready-to-run script. Everything is computed in your browser from data/files.json — nothing is sent anywhere.

Recipes

Patterns that work on a login node or in a job script.